融合改进K-SMOTE与LightGBM算法的故障检测方法

打开文本图片集

中图分类号：TP18；P756.1 文献标识码：A 文章编号：2096-4706（2025）15-0174-05

Fault Detection Method Integrating Improved K-SMOTE and LightGBM Algorithm

BAOJiali

（SchoolofBigDataandArtificial Intelligence，ChengduTechnological University，Chengdu61173o，China）

Abstract： To addressthe issues of weak model generalization and insuffcient recognition accuracy for minority-class faultsausedbylass-imbalanceddatainidustrialfultdetection，tispaperproposaclasificationframewrktatintegates clustering optimizationand EnsembleLeaming，aiming to improve therobustness and acuracyof multi-classfault detection. It adoptsanimprovedSMOTEalgorithmbasedonK-meansclustering，nhances thedistributionrationalityofminority-clas samples through intra-clustersample interpolationstrategies，dynamicallydividesbalanceddatasetscombined withrandomized K-foldcros-validatonndconstructsanintegratedclasificaionmodeltraedbytheLightGBMalgorithm.Expeentson the UCI steelplate defectdataset showthatthe proposed methodachievesanaccuracyprecision，andrecallof093，0.920，and 0.920，respectively，significantlyoutperforming modelssuch asRandomForest andMLP.TheAUCvalues ofROCcurves forall categoriesarehigherthanO.98，withthehighestreaching1.0o，verifingitshighdiscrminativeabilityforcomplexfaultpatts.

Keywords： industrial fault diagnosis;K-SMOTE algorithm; LightGBM

0 引言

工业设备的智能化故障诊断是保障现代制造业安全运行、降低维护成本的核心技术之一。（剩余7568字）